SERIES
PAPER
DISCUSSION
IZA DP No. 6883
What Determines Religious School Choice?
Theory and Evidence from Rural Bangladesh
M. Niaz Asadullah
Rupa Chakrabarti
Nazmul Chaudhury
September 2012
Forschungsinstitut
zur Zukunft der Arbeit
Institute for the Study
of Labor
What Determines Religious School Choice?
Theory and Evidence from Rural Bangladesh
M. Niaz Asadullah
University of Reading,
SKOPE, University of Oxford and IZA
Rupa Chakrabarti
University of Reading
Nazmul Chaudhury
World Bank
Discussion Paper No. 6883
September 2012
IZA
P.O. Box 7240
53072 Bonn
Germany
Phone: +49-228-3894-0
Fax: +49-228-3894-180
E-mail: [email protected]
Any opinions expressed here are those of the author(s) and not those of IZA. Research published in
this series may include views on policy, but the institute itself takes no institutional policy positions.
The IZA research network is committed to the IZA Guiding Principles of Research Integrity.
The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
organization supported by Deutsche Post Foundation. The center is associated with the University of
Bonn and offers a stimulating research environment through its international network, workshops and
conferences, data service, project support, research visits and doctoral program. IZA engages in (i)
original and internationally competitive research in all fields of labor economics, (ii) development of
policy concepts, and (iii) dissemination of research results and concepts to the interested public.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available directly from the author.
IZA Discussion Paper No. 6883
September 2012
ABSTRACT
What Determines Religious School Choice?
Theory and Evidence from Rural Bangladesh *
This paper looks at the determinants of school selection in rural Bangladesh, focusing on the
choice between registered Islamic and non-religious schools. We consider a two period
framework where children are a source of old age transfers. The amount of old age transfers
made by children as adults is influenced both by their schooling and by parental religiosity.
Parents also derive utility from educating a proportion of their children in madrasahs in a way
that reflects their own religious values. We investigate how household income, religious
preferences, schooling costs, and school quality affect the proportion of children sent to each
school type. Using a unique dataset on secondary school age children from rural
Bangladesh, we find that madrasah enrolment falls as household income increases. At the
same time, more religious households, and those that live further away from a non-religious
school are more likely to send their children to madrasahs. However, in contrast to the
theory, we find that Islamic school demand does not respond to the average quality of
schools in the locality.
JEL Classification:
Keywords:
D04, I21, O15
school choice, madrasah education, Bangladesh
Corresponding author:
M. Niaz Asadullah
School of Economics
University of Reading
PO Box 218
Whiteknights
Reading, RG6 6AA
United Kingdom
E-mail: [email protected]
*
This paper is an output of the International Growth Centre (IGC). Niaz Asadullah gratefully
acknowledges additional support from the Leverhulme Trust UK and the hospitality of the Research
and Evaluation Division (RED) of BRAC, where the final version of this paper was prepared. We are
grateful to Nigar Hashimzade and Mark Guzman for guidance and advice and Danny Cohen-Zada for
an informed discussion of the issues surrounding the paper. We would like to thank Zihad Hassan who
managed the survey work with utmost professional competence. Lastly, we would also like to thank
seminar participants at Reading University, participants at the 2011 Association for the Study of
th
Religion, Economics and Culture (ASREC) Conference, 11 UKFIET International Conference on
Education and Development (September 2011) and IGC seminar in Dhaka (March 2012) for their
helpful comments on an earlier version of the paper. The usual disclaimers apply.
1.
Introduction
For geo-political and international security related reasons, there is growing global
interest in Islamic faith schools or madrasahs. In Bangladesh and other South Asian
countries such as India and Pakistan, these schools allegedly cater to a large number of
students and thrive in poor and isolated communities. Enrolment in madrasahs, it is
argued, is on the rise. Madrasahs may promote extreme political and religious viewsand
gender norms and arguably do not impart significant marketable skills (Asadullah and
Chaudhury, 2010; World Bank, 2010). Therefore it begs the question why Muslim
households choose to send their children to these schools.
In a country with nearly 40% of the population living below the poverty line, household
choice of religious schooling in Bangladesh is popularly perceived in terms of poverty
and an inadequacy of mainstream schools (Evans, 2008). The weight of the evidence from
existing research indicates that non-religious 1 (government owned and financed) schools
are of better quality (Asadullah, Chaudhury and Dar, 2007) so that children are likely to
enrol in madrasahs only in areas where there is insufficient provision of non-religious
schools. Nonetheless, enrolment decisions can favour madrasahs owing to economic
considerations even in better provided areas. Although many madrasahs are fee-charging,
they present a cheaper alternative to non-religious schools (World Bank, 2010) and are
hence likely to appeal to relatively poorer households. The quality of non-religious
schools in rural locations may not be sufficiently high to attenuate the overall demand for
madrasahs. However, these hypotheses remain untested given limited research on parental
choice of schools in Muslim countries 2.
Religious school choice is however a relatively well researched topic for developed
countries. Several European countries such as Ireland, Netherlands, and the United
Kingdom have a large number of state-financed religious schools, whilst the United States
has a significant number of private schools that operate under a religious orientation. The
extant literature on faith school choice in these countries is suggestive of a variety of
reasons for demand for religious education (Sander, 2005; Cohen-Zada and Justman,
1
The term “non-religious” indicates that the school admits students from all faith groups.
There are three recent studies on faith schools in developing countries -Wodon and Ying (2009) and
Backiny-Yetna and Wodon (2009a, 2009b). But none of these focuses on the question of school choice –
they only present evidence on the relative performance of different school types.
2
2
2005; Cohen-Zada, 2006; Cohen-Zada and Sander, 2008). Cohen-Zada (2006) models the
decision of religious parents to send their children to private religious schools in terms of
parental desire to protect their children from external influences and thus preserve their
religious identity. Using data on Catholic private school enrolment in the US, he finds that
when the share of the religious minority in the local population grows, and outside
influences become less threatening, the demand for separate religious schooling among
the members of the religious group decreases. Sander (2005) finds that the probability of
children attending Catholic schools is explained by Catholic parents’ religiosity (proxied
by church attendance). In a more recent study, Cohen-Zada and Sander (2008) quantify
the religious factor in private education in the US by estimating a random utility model of
school-choice in which households choose among public, private/non-sectarian, Catholic
and Protestant schools. They find that both religion and religiosity have important effects
on the demand for Catholic schools 3.
Some of the above results can help shed some light on religious school choice in rural
Bangladesh. Religious preferences may be important in explaining madrasah enrolment,
but at the same time, household choices can be driven by economic costs and benefits. In
this latter respect, the household decision for a family in rural Bangladesh is not dissimilar
to that of a family in a developed country. Irrespective of their mode of finance and
management, faith schools (Catholic and/or Church affiliated) in the US are considered to
be superior to public schools in terms of quality (for example see Neal, 1997). In the US,
these faith schools are overwhelmingly under private management and charge high fees
which limit participation of children from poorer socio-economic backgrounds. Public
school enrolment is thus inversely correlated with household income. In Bangladesh, it is
the religious schools which offer a cheaper, but lower quality alternative to non-religious
schools. In this sense, parents in both settings face a trade-off between school cost and
school quality4.
3
However, in a companion paper, Cohen-Zada and Sander (2008b) relax the assumption that religiosity
(attendance at religious services) is an exogenous determinant of religious school choice.
4
Research on household school choice in developing countries also confirms the relative importance of
quality. For instance, Alderman, Orazem and Paterno (2001) report that the poorest households in Pakistan
send children to fee-charging private schools (and that utilization increases with income) when public
schools are of poor quality.
3
For historical reasons, Bangladesh was endowed with a relatively large governmentregistered madrasah system (Asadullah and Chaudhury, 2009a; World Bank, 2010) 5.
These madrasahs are recognised by the state and follow a standardized curriculum which
also includes subjects such as mathematics, languages and science, in addition to Islamic
studies. This distinguishes Bangladesh from other countries in South Asia (where
madrasahs are typically unrecognised by the state and focus almost exclusively on
religious instruction), and presents an ideal context in which to study household choice
between registered madrasahs and non-religious schools. So far, this question has not
been addressed empirically owing to the lack of household survey based information on
enrolment in Islamic religious schools.
In this paper, we use data from a purposefully designed rural household survey QSSMEB (Quality of Secondary School Madrasa Education in Bangladesh), which
examines the incidence of madrasah education in Bangladesh, and households’ decisions
to educate children in madrasahs. The QSSMEB survey was designed to present definitive
statements regarding the choice of madrasahs among households as a schooling choice.
Therefore, we go beyond simple tabulations and present regression estimates of
determinants of madrasah enrolment. The regression specification is motivated by
predictions from a two-period model of school choice.
One of the interesting features of school enrolment in rural Bangladesh is that a household
may educate some of its children in a madrasah while the others attend a non-religious
school. In the QSSMEB data, nearly 70% households with at least two secondary school
age children chose madrasah education for some children and school education for the
others. We thus develop a simple two period framework to explain how a household
chooses the proportion of children to send to each school type. The model treats children
as a source of old age transfers where the amount of transfers is determined by the labour
market outcome of the children as well as their traditional and religious values 6. These
5
The first registered madrasah, initially known as the Calcutta Madrasah, was established in 1781 in
Calcutta. The madrasah taught both religious and secular subjects, and was supposed to be a model for
madrasah reform for the rest of India. After the India-Pakistan partition in 1947, it was transferred to Dhaka.
This encouraged significant growth in the number of registered madrasahs throughout East Pakistan. In the
early 1980s there was a further proliferation of Aliyah madrasahs in response to a series of reform measures
initiated by the government.
6
There is experimental evidence from South Asia which suggests a link between religiosity and pro-social
behaviour. For instance, in a field study conducted in India, Ahmed (2009) finds that madrasah students are
4
values are in turn influenced by their parents’ religiosity as well as their schooling.
Parents also derive satisfaction from sending a proportion of their children to madrasahs
in a way that reflects their own religious values. Keeping in mind that children who are
madrasah graduates are likely to earn less as adults, parents need to balance the economic
motives of educating children with the non-economic motives of school choice which are
driven by their personal beliefs. The model then examines how factors such as household
income, religious preferences, schooling costs and school quality affect the proportion of
household children educated in each school type.
By way of combining economic theory with field-based evidence, our paper makes a
unique contribution to the current international literature on madrasah education. It builds
on two closely related studies on the issue, namely, Newhouse and Beegle (2006) and
Andrabi, Das, Khwaja, and Zajonc (2006). Newhouse and Beegle present a model to
explain household choice of a school type based on their wealth and preference for
academic achievement. Whilst their analysis is a useful starting point, they do not
explicitly focus on religious school choice. Instead, they focus on the decision to send
children to private (religious or non-religious) or public schools in Indonesia.
On the other hand, Andrabi, Das, Khwaja, and Zajonc (2006) focus on the question of
religious school enrolment and present descriptive evidence on the issue for Pakistan. The
authors find that school choice by households in their data appears to be driven neither
solely by poverty, nor by the religious mindedness of households. On the basis of
household censuses and surveys, it is shown that one household could send one of its
children to a madrasah, another to a public school, and a third to one of the increasingly
common private schools. Nonetheless, the households that choose madrasahs account for
a very small proportion in the study sample which limits the scope for a formal statistical
analysis of the issue. Another study of school choice in Pakistan is carried out by
Alderman, Orazem and Paterno (2001); however, they do not address the issue of
religious school choice per se. In their study, the focus is on household choice between
private schools, public schools, and not sending children to school at all.
significantly more cooperative in the public goods game and significantly more generous in the dictator
game than other students.
5
Research on school choice in Muslim communities can be complicated by the fact that the
institution of madrasah varies widely in terms of what is taught, how it is taught, and what
view of religion is imparted to students. In addition to the large number of registered
madrasahs that operates with complete oversight of the government, there is an unknown
number of madrasahs in Bangladesh which are not recognised by the state. This latter
group primarily specializes in religious education and does not teach marketable skills.
Therefore, unrecognised madrasahs do not serve as a substitute for formal schools in the
labour market. Moreover, the enrolment share of these madrasahs in rural Bangladesh is
very small (World Bank, 2010). For these two reasons, this study does not explore the
question of unrecognised madrasah attendance.
The rest of the paper is organized as follows. Section 2 presents background information
on madrasahs and the mainstream education system in Bangladesh. Section 3 presents a
simple theoretical model of school choice. Section 4 discusses the dataset whilst empirical
results are presented in Section 5. Section 6 concludes.
2. Background
Islamic religious schools in Bangladesh are almost entirely in the non-state sector,
operating at both primary and secondary levels. Moreover, the majority of the staterecognised schools also benefit fiscally from the state under the Private-Public Partnership
(PPP) model. Little systematic information is available on those religious schools that
remain outside the state sector and are unrecognised by the state. There is a perceived
difference among various madrasahs in terms of in-school conduct and curriculum
content. Indeed, madrasahs operating at primary or secondary levels in Bangladesh are
classified in two broad categories: Aliyah madrasahs and Quomi madrasahs. Outside the
primary/secondary sector 7, there is a large number of pre-primary Islamic education
institutions which are of two types, Maktab (or Nourani madrasah) and Furqania/Hafizia
madrasahs 8. In this study, we focus on secondary education choices and outcomes only.
7
As in other South Asian countries, the Bangladeshi education system consists of four main levels: primary
school (five years of study), secondary school (five years), and higher secondary education (two years) and
tertiary (graduate and post-graduate) education.
8
Abdalla, Raisuddin and Hussein (2004) provide an excellent discussion of the curriculum and structure of
primary and pre- primary religious and non-religious schools within the state regulated sector.
6
The establishment of Aliyah madrasahs and their activities follow government regulations
as prescribed by the Madrasah Education Board. The Board approves curriculum for all
registered madrasahs from secondary to post-graduate level. For a madrasah operating at
any level, there are clearly prescribed requirements for state recognition and support. In
terms of curriculum content, Aliyah madrasahs offer a hybrid education where students
are taught both religious subjects and modern general education.
The secondary level of the Aliyah stream is known as Dakhil which spans grades 6 to 10.
As inthe mainstream education system, at the end of Dakhil education students appear in a
public examination which is organized by the Bangladesh Madrasah Education Board.
The Dakhil certificate is equivalent to the Secondary School Certificate (SSC) obtained
after successful completion of secondary education from non-religious schools. There are
four Dakhil madrasah curriculum groupings: (a) General group, where the focus is on
languages, mathematics, religious studies, and social studies; (b) Science group, where in
addition to languages, mathematics, and religious studies, students are examined in
computer science, physics, chemistry, and biology; (c) Muzabbid group, where students
are taught in additional religious subjects such as Quran, hadith, and fiqh; (d) Hifzul
Quran group, where students specialize in memorizing the Quran.
Similarly, students in non-religious secondary schools can opt for different group
specializations such as science, arts and commerce. However, they cannot specialise in
religious studies; at most 10% of the curriculum comprises of Islamic education 9.
Therefore, in comparison to mainstream non-religious schools, Aliyah secondary
madrasahs give students the option to specialise in general, non-religious stream as well
as Islamic theology.
The international discourse uses the word “madrasah” as a catchall term. However,
depending on level of education and/or government recognition, madrasahs differ in
types. First, a madrasah can mean something as simple as a pre-school institution, where
children learn the Quran alongside a few religious practices. Second, it can mean a
primary or secondary school that serves as a substitute for mainstream non-religious
education. Third, it can also mean a seminary established to train proper clerics in
9
Non-Muslim students study their own religion or are allowed to opt out of religious education.
7
classical Islamic religious knowledge. We, therefore, define madrasah enrolment by
types.
World Bank (2010) presents the market share of different types of madrasahs in rural
Bangladesh using the QSSMEB survey data which was collected in 2008. According to
the QSSMEB Census, there are 32,372 children between the ages of 6 and 18 of which
17.52% (i.e. 5763 children) report the last school attended as a madrasah. However, many
students may have first enrolled in a madrasah but have subsequently moved to a school.
When accounting for this fact, 19.78% (6404 children) have enrolled at least once in
madrasahs in the past. Lastly, when restricted to the population of 24,440 children who
are currently enrolled, the share of madrasahs rises to 18.79%.These numbers are
consistent with household survey based estimates of the market share of madrasahs.
Table 1 below summarizes data on the market share of madrasahs by levels of education.
The share is found to be much higher in secondary education. These figures are close to
estimates available from other sources. For instance, according to the large scale nationwide household survey of CAMPE (Campaign for Popular Education), madrasahs account
for 16% of total enrolment in the primary sector whilst our estimates range between
13.8%and 15.7%. Unfortunately, apart from CAMPE, there is no other source of
comparable data on madrasah enrolment.
Table 1: Madrasah enrolment in rural Bangladesh
Overall
share
Breakdown by madrasah type
QSSMEB 2008 census(24841 households)
% share of primary enrolment
13.8%
8.14% is in registered madrasah and 5.4% is in
unregistered madrasah
% share of secondary enrolment
22.1%
18.5% is in registered madrasah and 3.6% is in
unregistered madrasah
% share of primary enrolment
15.7%
9.48% is in registered madrasah and 5.9% is in
unregistered madrasah
% share of secondary enrolment
24.5%
20.3% is in registered madrasah and 4.2% is in
unregistered madrasah
QSSMEB 2008 survey(2400 households)
8
Finally, differences in curriculum aside, registered schools and madrasahs differ
significantly in terms of costs and physical inputs. Table 2 reports the average profile of
non-religious secondary schools and registered madrasahs in rural Bangladesh. On
average, schools have a bigger land area, are better inspected by the government and have
a larger proportion of trained teachers. They also charge higher fees and pay a higher
salary to their teachers. Looking at average test scores in mathematics and English, we
find that non-religious schools perform slightly better, while as expected, madrasahs have
a clear advantage when it comes to religious studies. However, it is worth noting that
while non-religious schools tend to achieve somewhat higher scores, the average
performance is rather poor for both types of schools.
Table 2: Profile of an average non-religious secondary school and madrasah
Non-religious
school
School land area (in logs)
4.91
School inspected(1 if yes)
0.82
Student fees (in logs)
4.95
% of female teachers
18
% of trained teachers
69
Teacher salary (in logs)
8.63
Grade 8 student score in maths (% of answers correct)
35.57
Grade 8 student score in English (% of answers correct)
36.88
Grade 8 student score in Islam (% of answers correct)
69.94
N
206
Source: Based on World Bank (2010) and QSSMEB survey 2008.
Registered
Madrasah
4.77
0.69
4.20
9
25
8.29
33.14
26.69
81.54
123
3. A theoretical model of household demand for madrasah education
As pointed out earlier, religious school choice has not been studied extensively in
developing, and in particular, Islamic countries. However, one example is Newhouse and
Beegle (2006) who address the issue for Indonesia where households have the option of
sending their children to private or public (religious or non-religious) schools. To
formalize the problem of school choice, they set up a two-period model where parents
select a school in Period 1, whilst the child transfers an amount of income determined by
the quality of their education, which is consumed by the parent in the second period.
Parents’ utility is a positive function of their consumption in the two periods, their taste
for school quality (or valuation of academic achievement) as well as non-academic
9
characteristics of the school such as safety, religious affiliation, convenience and
discipline.
The Newhouse and Beegle model makes two predictions. First, a parent that places a
higher value on education is more likely to send their child to public schools, which are
assumed to be of higher average quality. Second, holding other factors constant, wealthier
households are more likely to send their children to private school.
In Bangladesh, most of the non-religious schools are either state managed or aided. The
share of the unrecognised madrasahs in secondary education is insignificant. Therefore,
the real choice lies between state-recognised secondary schools and madrasahs, rather
than between public and private faith schools. In this paper, we develop a theoretical
model to explain the household choice of madrasahs over non-religious schools in rural
areas. Unlike in the Newhouse and Beegle model where religious affiliation of the school
is considered a part of “other” factors, we specifically model both the impact of the
religiousness of the school curriculum as well as parental religiosity on the level of
transfers. Parents also derive direct satisfaction from their school choice decision relative
to their religiosity. In addition, our model specifically addresses the observed fact that a
proportion of children in a given household might attend madrasahs while the rest attend
non-religious schools.
We consider a simple, two-period model, where individuals work when young, earn an
exogenous income, y, and send their children to school 10. Schools are of two types, nonreligious secondary schools, which are denoted by s and state recognised Aliyah
madrasahs, denoted by m.
Schooling costs, denoted by Ps and Pm respectively, include both tuition and travel cost 11.
Empirical evidence suggests that tuition for registered Aliyah madrasahs is typically less
than that of non-religious schools, and they are also likely to be geographically more
10
For the purposes of this model, we only focus on households that send their children to school. We are
thus abstracting away from the issue of child labour, whereby some parents might decide not to send their
children to school at all. In our sample, 72% of secondary school age children attend a secondary school of
some type.
11
This is done for simplicity. We could split up total schooling cost into 𝑃𝑗 = 𝑄𝑗 + 𝐷𝑗 , where 𝑄𝑗 represents
tuition, while 𝐷𝑗 represents distance or travel cost. Although tuition in secular schools is higher, we can
allow for the travel cost to the madrasah to be higher or lower than travel cost to a non-religious school,
depending on the location of schools in a particular village.
10
accessible than secondary schools; thus 𝑃𝑠 > 𝑃𝑚 . We further assume that both Ps and Pm
are strictly less than y, and that 𝑦 − 𝑃𝑚 > 𝑦, where 𝑦 is a subsistence level of income.The
number of children is normalized to 1. Parents send a proportion a of their children to
madrasahs, and a proportion 1 – a to non-religious schools.
In the second period, parents get old age transfers from their children. For simplicity, we
assume that there is no other income in period 2; however, adding this will not make a
significant difference to the analysis. Transfers (T) are a proportion of the wage (w)
earned by the child as an adult. This proportion is determined by their traditional and
religious values as an adult. In the case of rural households, growing up in a more
religious home and attending a more religious school, is likely to instil stronger traditional
and religious values in a child. One aspect of these traditional values is looking after
parents in their old age. Let 𝑍𝑗 be an index of these adult values for a child who was
educated in school type j. Thus,
𝑇𝑗 = 𝑍𝑗 𝑤𝑗 , wherej = s, m
(1)
A child’s adult wage is determined by the type of school its parents chose in period 1.
Aliyah madrasahs teach fewer marketable skills, therefore, 𝑤𝑠 > 𝑤𝑚 12.We can also think
of these wages as a proxy for school quality, although it is worth keeping in mind that in
some rural areas, school quality may be low across the board (as seen in Table 2).
The traditional and religious values of a child in adulthood, 𝑍𝑗 , is an increasing function
of both the religious aspect of their school curriculum, as well as their parent’s religiosity:
Thus,
𝑍𝑗 = 𝑍𝑗 �𝑧𝑗 , 𝑍𝑃 �
(2)
where𝑧𝑗 represents the religious aspect of the curriculum in school type j (since madrasahs
have a more religious curriculum, 𝑧𝑚 > 𝑧𝑠 ), and 𝑍𝑃 is the parent’s religiosity, which is
taken as given in this two period model. In general, 𝑍𝑠 and 𝑍𝑚 can have different forms.
12
See Asadullah (2009) for evidence on the negative correlation between wages and madrasah attendance.
11
Finally, the total amount of transfers received by the parent in period 2 is a weighted
average of transfers from madrasah-educated and non-religious school-educated children,
and is given by
𝑇 = 𝑎𝑇𝑚 + (1 − 𝑎)𝑇𝑠
(3)
Apart from consumption in the first and second period (𝑐1and 𝑐2 , respectively), parents
also derive utility from sending a certain proportion of their children to madrasahs in a
way that reflects their religious values. The lifetime utility of the representative parent is
given by
𝑎
𝑈 = ln 𝑐1 + 𝛽 ln 𝑐2 + 𝛾𝑣 �𝑍 �
𝑃
(4)
where 𝑣 ′ > 0 if (𝑎⁄𝑍𝑃 ) < 1
𝑣 ′ = 0if(𝑎⁄𝑍𝑃 ) = 1
𝑣 ′ < 0if (𝑎⁄𝑍𝑃 ) > 1, and
𝑣 ′′ < 0for all (𝑎⁄𝑍𝑃 ).
0 < 𝛽 < 1, and
𝛾>0
We further assume that 𝑣(0) = 0, and that there is a lower bound on parental religiosity,
𝑎
𝑍𝑃 , such that 𝑣 �𝑍 � = 0. This is graphically illustrated in Appendix 1.
𝑃
The first two terms of the utility function are the standard consumption terms, while the
last term in the utility function can be thought of as a “school choice” term. This school
choice term in Equation (4) says that sending a certain proportion of children to
madrasahs yields the highest utility when this proportion best matches the religious values
of the parent, i.e. when (𝑎⁄𝑍𝑃 ) = 1. A parent who is not very religious, but still sends a
large proportion of children to a madrasah (because it is the cheaper or more accessible
alternative), gains lower utility from the school choice term, since in their case, (𝑎⁄𝑍𝑃 ) >
1. Similarly, for a parent who is very religious, but does not send a large enough
proportion of children to a madrasah, the additional benefit from the school choice term is
not very high.
12
The utility function in effect says that parents want their children to be similar to
themselves in their religious values. Parents thus try to match their school choice (a),to
their own religiosity (𝑍𝑃 ). However, religious parents do not simply send all their children
to madrasahs. The religious motivation for school choice is only part of the story. Parents
also care about their consumption in both periods, which, as the budget constraints below
show, depend on income, school costs, and the potential transfers they get from their
children in old age.
The problem for the representative parent is to choose a, the proportion of children to
send to madrasah, in order to maximise lifetime utility.
𝑎
max ln 𝑐1 + 𝛽 ln 𝑐2 + 𝛾𝑣 � �
𝑎∈[0,1]
𝑍𝑃
subject to,
𝑐1 = 𝑦 − 𝑎𝑃𝑚 − (1 − 𝑎)𝑃𝑠
𝑐2 = 𝑎𝑇𝑚 + (1 − 𝑎)𝑇𝑠
It can be readily verified that the first order condition for this problem is:
(𝑇𝑠 − 𝑇𝑚 )
𝑃𝑠 − 𝑃𝑚
𝑎
1
−𝛽
+ 𝛾𝑣′ � � � � = 0
𝑐2
𝑐1
𝑍𝑃 𝑍𝑃
(5)
We can now substitute equations (1) and (2) into equation (5) to express the first order
condition in terms of the different parameters of the model. In particular, we can express
the first order condition as:
𝐹(𝑎, 𝑦, 𝑃𝑠 , 𝑃𝑚 , 𝑤𝑠 , 𝑤𝑚 , 𝑧𝑠 , 𝑧𝑚 , 𝑍𝑃 ) = 0
(6)
Using the implicit function theorem 13, we can now see how the optimal proportion of
children educated in madrasahs, varies with changes in school costs, school quality (as
13
The effect of each parameter on a, the proportion of children enrolled in madrasahs, can be calculated as
follows; using the parameter y as an example, we know that 𝜕𝑎 ⁄𝜕𝑦 = −�𝐹𝑦 ⁄𝐹𝑎 �.
13
proxied by future wages), the religiousness of the school curriculum, and parental
religiosity.
Comparative statics
From equation (6), it can be shown that as parental income y increases, the proportion of
children enrolled in madrasahs decreases. This is because an increase in y means that
parents can afford the more expensive non-religious schools and thus they educate a
smaller proportion of their children in madrasahs 14.
As madrasahs become more
expensive, i.e. as the cost of madrasahs (𝑃𝑚 ) increases, madrasah enrolment falls.
Similarly, if non-religious secondary schools become more expensive, then madrasah
enrolment rises.
From equation (6), it can also be shown that if either 𝑤𝑠 or 𝑧𝑠 increases, then madrasah
enrolment decreases. This is because in both cases, the parameters have a positive impact
on the amount of transfers (𝑇𝑠 ) that parents can expect in their old age from their non-
religious school educated children. Similarly, if either 𝑤𝑚 or 𝑧𝑚 rises, then madrasah-
educated children become more valuable as a source of transfers and thus madrasah
enrolment also rises.
Finally, we look at the impact of parental religiosity, 𝑍𝑃 . What happens to madrasah
enrolment as parental religiosity changes now depends on how the child’s religiosity as an
adult, 𝑍𝑗 , responds to parental religiosity, 𝑍𝑃 . It will be useful to define the “elasticity” of
children’s religiosity with respect to their parent’s religiosity as:
𝜀𝑗 =
𝜕𝑍𝑗 ⁄𝜕𝑍𝑃
𝑍𝑗 ⁄𝑍𝑃
(7)
Using the implicit function theorem, we know that
14
It can be shown that if parents earn income in both periods, then a depends negatively on first period
income, as shown here, and positively on second period income. This is not surprising since parents who
have a larger second period income do not need to rely as much on transfers, and hence can afford to
educate a larger proportion of their children in madrasahs.
14
𝐹𝑍
𝜕𝑎
=− 𝑃
𝐹𝑎
𝜕𝑍𝑃
𝜕𝑎
From equation (6), it can be easily verified that 𝐹𝑎 < 0. Thus, the sign of 𝜕𝑍 is
determined by the sign of 𝐹𝑍𝑃 , which is given by the following equation:
𝑃
𝛽𝑤𝑚 𝑤𝑠 𝑍𝑚 𝑍𝑠
𝑎
𝑎 1
𝑎 1
(𝜀𝑚 − 𝜀𝑠 ) − 𝛾𝑣" � � � 2 � − 𝛾𝑣′ � � 2
2
𝑍𝑃 𝑍𝑃 𝑍𝑃
𝑍𝑃 𝑍𝑃
𝑍𝑃 𝑐2
𝐹𝑍𝑃 =
(8)
Assuming that (𝜀𝑚 − 𝜀𝑠 ) ≥ 0 15 and given that 𝑣" < 0, we can see that the first two terms
15F
in equation (8) are positive. For values of (𝑎⁄𝑍𝑃 ) > 1, we know that 𝑣 ′ < 0, which
implies that the last term in equation (8) is also positive. Furthermore, when (𝑎⁄𝑍𝑃 ) = 1,
we know that 𝑣 ′ = 0. Thus the sign of 𝐹𝑍𝑃 is positive for the region where (𝑎⁄𝑍𝑃 ) ≥ 1.
𝜕𝑎
In other words, 𝜕𝑍 > 0 for the region where (𝑎⁄𝑍𝑃 ) ≥ 1; thus madrasah enrolment
𝑃
increases with an increase in parental religiosity.
𝜕𝑎
𝑎
For the region where (𝑎⁄𝑍𝑃 ) < 1, the sign of 𝜕𝑍 remains positive until a point �𝑍 �
𝑃
′
𝑃
𝑙𝑜𝑤
below which 𝑣 has a sufficiently large positive value so as to outweigh the effect of the
𝜕𝑎
first two terms in equation (8). In this case, we find that𝜕𝑍 < 0.
𝑃
Thus, in general, parental religiosity has a positive effect on madrasah enrolment.
However, there is a very small range of values of (𝑎/𝑍𝑃 ), for which this effect might be
reversed. This occurs where (𝑎/𝑍𝑃 ) is close to zero. For people who are very religious
and for whom 𝑍𝑃 is very large relative to a, the utility from the school choice term in
equation (4) is very small, and it is the consumption effect that dominates. Thus if they
were to become even more religious, they would need to compensate for their reduction in
Generally speaking, 𝜀𝑚 can be greater than or less than 𝜀𝑠 . However, it is quite plausible that the
religiosity of madrasah-educated children is more responsive to parental religiosity. Madrasahs, with their
greater emphasis on religious instruction, are more likely to impart traditional values than non-religious
schools, and this includes being more sensitive to one’s parent’s religiosity. Consequently, we henceforth
assume that (𝜀𝑚 − 𝜀𝑠 ) ≥ 0.
15
15
second period consumption by increasing the proportion sent to non-religious schools
(keeping in mind that non-religious school graduates will earn a higher wage, and thus
give higher transfers than their madrasah educated counterparts). Table 3 below
summarises how a responds to changes in various parameters of the model.
Table 3: Summary of comparative statics
Parameter
Parental income (y)
Cost of madrasah(𝑃𝑚 )
Cost ofnon-religious schools(𝑃𝑠 )
Adult wage of secondary-educated child or non-religious
school quality (𝑤𝑠 )
Adult wage of madrasah-educated child or madrasah quality
(𝑤𝑚 )
Religiousness of non-religious curriculum (𝑧𝑠 )
Religiousness of madrasah curriculum (𝑧𝑚 )
Parental religiosity (𝑍𝑃 )
Effect on the fraction of children sent
to madrasahs
+
+
+
+/-
We now move on to the empirical part of the investigation. Due to the nature of the
survey data, it was not possible to match up every single parameter from the theoretical
model to the empirical study. The comparative statics presented in Table 3 are thus used
as a guide for specifying the school choice equation for the regression analysis.
4. Data
The empirical analysis presented in this paper is based on a unique survey of households
carried out by the authors in 2008, as part of a project on religious school choice in rural
Bangladesh. The survey was conducted in 96 villages. The village sample frame
comprised of all households where there is at least 1 child aged between 11 and 18 years.
The census, on the other hand, covers all households in the sample villages in 12 districts,
spread across 6 divisions in Bangladesh. Consequently, it yields a much bigger sample on
madrasah enrolment which allows for examining correlations with household attributes in
a meaningful manner.
In addition to directly asking the question: “Is the child enrolled in an Islamic education
school?”, the QSSMEB survey questionnaire also enquired about any madrasah/school
16
the child may have attended previously. The survey questionnaire also has a detailed
module on past school choice for all children of school age. Lastly, schooling information
is collected not only for resident children, but also for children of the head of household
who currently do not live in the household. This is an important issue, since many
madrasahs in South Asia are boarding schools, so that ignoring non-resident children
would lead to an under-estimate of the true enrolment share of madrasahs.
For the above reasons, we are able to define madrasah enrolment in broad terms – we go
beyond current enrolment by using data on “ever enrolment” and use information on
resident as well as non-resident children of school age. Another peculiarity of Bangladesh
is a sizeable non-Muslim population. However, since madrasahs cater exclusively to the
Muslim population, we discard all non-Muslim households in our empirical analysis.
Lastly, although completion rates still remain poor, secondary school participation rate in
Bangladesh has increased considerably in recent years. As a proportion of all secondary
school-age (i.e. 11-18 years old) children, 72% children in our data were enrolled in
secondary education. Since we are interested in explaining the proportion of secondary
school going children sent to registered madrasahs, we discard data on the small fraction
of households where none of the children have ever attended secondary school.
Incorporating this small sub-group of households in our analysis requires joint modelling
of schooling and child labour decisions. Since our primary objective is to explain
household’s choice of school type with a focus on religious schooling, we have abstracted
away from the problem of how to decide whether to send the child to secondary school or
not in the first place.
5. Findings
We begin by describing the variation in madrasah enrolment at the level of the household
in the QSSMEB survey data. We follow Andrabi et al. (2006) and analyse enrolment
choices at the household level. First, for households that have a child enrolled in a
madrasah, we examine what type(s) of schools the other children in the household attend.
Similarly, we repeated this exercise in households with at least one child going to a nonreligious school. Lastly, focusing on households that had at least two enrolled children we
classify them as (i) “all madrasah” if all children attend registered madrasah; (ii) “mixed,”
17
ifthe households had some children in school and others in madrasahs simultaneously and
(iii) “all school,” if all children attend school.
The results are summarized in Table 4. Among households with at least one child enrolled
in a school (i.e. “school households”), 76% of households choosing schools for at least
one of their children select non-religious schools for all of their school-age children, while
18% of households sending a child to a madrasah (i.e. “madrasah households”) send all of
their children to madrasahs. In other words, the majority of “madrasah households” use
both madrasahs and non-religious schools whilst the pattern is the opposite in case of
“school households”. Among households with at least two children enrolled in secondary
education, 5% can be classified as “all madrasah” households. Therefore, a small but
significant part of the variation in madrasah enrolment is between households and lends
some support to the hypothesis that the choice of a madrasah in Bangladesh provides
some information about the ideology of the household.
Table 4:Household enrolment choices
Households
Households
All
with at least 1 with at least 1
households
school
madrasah
enrolled child enrolled child
% Children in non-religious school and madrasah
23.64
81.69
22.45
% Children in madrasah only
18.31
5.03
% Children in non-religious school only
76.36
72.52
(N)
(1472)
(426)
(1550)
Note: Figures exclude children enrolled in unregistered madrasahs and based on children aged 6-18 years.
The fact that a significant part of the variation in madrasah enrolment is between
households highlights the role of household-specific factors in predicting madrasah
enrolment of children. Motivated by religious convictions, households may only have to
“sacrifice” one child for religious education. Indeed, analysis of household responses to
direct question on reasons for madrasah attendance does provide support for some of
these theories. The most commonly cited reason is religious preference of the household
(45.79%), followed by perceived high quality of madrasah education (22.18%), concerns
for after-life (8.22%) and distance to school (6.88%).
18
The significance of the variation in madrasah enrolment at the household level is also
confirmed by the descriptive statistics presented in Appendix Table 1. Further insights
into the underlying reasons for madrasah attendance can also be obtained by looking at
the figures presented in Appendix Table 1. The first column in the Table reports mean
values of key household and community-level variables. The second and third columns
report the mean values of the variables at the child level separately for school and
madrasah enrolled children respectively. In our sample, there are a total of 1566 children
enrolled in registered secondary school/madrasah of which 19% are reported to be
attending religious schools.
A number of observations can be made here. First, households whose children are
enrolled in non-religious schools have higher income and better access to electricity.
Second, madrasah attending children belong to more religious households. In the religious
households (defined as parents praying five times every day) sub-sample, 8.89% of the
households send all children to madrasa compared to a 3.48% in non-religious household
sample. However, even in the religious household sub-sample, 24% of all households
send one child to school and one to madrasah suggesting that in rural Bangladesh, both
religion and economics are important in explaining school choice. Third, madrasah
enrolled children live in less developed (measured in terms of average educational
attainment of the villagers and road conditions) and more religious (proxied by the
number of mosques) villages. Finally, the demand for madrasah schooling appears to be
inversely related to the quality of non-religious schools in the area (measured in terms of
test scores) – households with at least one madrasah enrolled child report living in areas
with higher average madrasah test score (mathematics as well as English). The opposite
pattern is noted for households with at least one non-religious school child. We now
further explore these issues in a regression framework.
Correlates of madrasah enrolment
In this section, we build on simple tabulations presented earlier and formally explore the
correlates of madrasah enrolment in a regression framework. We are interested in learning
about the effect of variables such as household poverty, parental education, religious
preference, distance to schools and community profile on school choice decisions. The
study of school choice is complicated by the presence of two types of religious schools.
19
Given the institutional differences between recognised and unrecognised madrasahs and
the very small share of the latter in total secondary school enrolment, we discard the
unrecognised ones from the regression analysis. The regression model is estimated using
data on all secondary school enrolled children, leaving non-religious school as the base
category type.
The regression sample contains 1229 households where 27% households have more than
1 secondary school enrolled children. Our dependent variable of interest is the proportion
of children belonging to the same household who attend madrasahs. Therefore, the
dependent variable has values that fall between zero and one. Estimation of the regression
model using Ordinary Least Squares (OLS) method with proportions response dependent
variable is unproblematic as long as all of the proportions lie well within the unit interval
(e.g. between values 0.3 and 0.7). Otherwise, the predicted values are unlikely to fall
between zero and one. In other words, in cases where the variable is strictly inside the unit
interval, one cannot recover the expected value of the fractional response from a linear
model for the log-odds ratio. Indeed, density plots of the dependent variable reveals a
significant number of proportions close to 1 and 0 (results not shown). Estimating a logit
model for such data is not appropriate either – using the log-odds transformation fails
when we observe responses at the corners, zero and one. This will result in missing
values, and hence observations would subsequently be dropped from the estimation
sample. Amongst other things, a linear functional form for the conditional mean might
also miss important nonlinearities. To deal with fractional response data, one needs
models for the conditional mean of the fractional response that keep the predicted values
in the unit interval. For the above reasons, Papke and Wooldridge (1996) suggest using a
Generalized Linear Model 16 (GLM), also known as Fractional logit, which can handle
proportions of exactly 0 or 1 17. However, for the sake of comparability, we also report
results from OLS regression 18.
16
The GLM generalizes linear regression by allowing the linear model to be related to the response variable
via a link function and by allowing the magnitude of the variance of each measurement to be a function of
its predicted value.
17
For a textbook treatment of fractional logit regression model, see Wooldridge (2002, page 661 p).
18
For empirical studies comparing estimates of OLS and Flogit, see Wagner (2001).
20
The choice of independent variables in our regression analysis is guided by the theoretical
model of school choice discussed in Section 3. The regression model has the following
controls:
a) household income (proxied by per capita expenditure; household assets; household
having electricity; mother’s and father’s education level and age 19)
b) parental religiosity
c) distance to different types of school from the household (measured in terms of
travel time, in minutes)
d) availability of registered madrasahs relative to non-religious secondary schools in
the locality (i.e. union)
e) quality of registered madrasahs and non-religious schools in the locality (i.e.
union)
f) average age and gender of school-age children in the household
g) village characteristics
This specification is similar to that used by Newhouse and Beegle (2006) 20. However, our
data is rich enough to shed light on the effect of poverty, household’s religious preference
and supply-side influence. Lastly, it should be pointed out that the model presented earlier
abstracts away from intra-household issues i.e. we did not focus on within household
variation in terms of factors that differ across children of the same household. It is wellknown in the literature on determinants of child outcomes that the demographic structure
of the household matters: children of certain birth order and gender enjoy systematic
advantage over their siblings in the same household (e.g. see Garg and Morduch, 1998).
In particular, poor households are budget constrained and that intensifies competition
amongst siblings. In this setting, children with certain demographic and possibly innate
attributes may be selected for religious education. Using this framework, one can explain
19
In addition, a positive coefficient on parental age may serve as a crude test of the “after-life consumption”
hypothesis. Using data on Taiwan, Chang (2005) finds that (a), religious giving and charitable giving are
closely related to after-life consumption and (b) older people are more likely to provide volunteer work and
attend more religious activities than younger individuals.
20
Their child-level regression model included the following variables: (a) household wealth (proxied by per
capita expenditure and floor type of the household); (b) past academic achievement of the child (including
elementary school grade repetition); (c) mother’s and father’s education level and religion; (d) student’s sex;
(e) whether the student worked in junior high; (f) the general size of the student’s village at age 12; (g) the
type of elementary school; (h) indicators for languages spoken at home; (i) the average district elementary
score; (j) number of schools in the district and (k) the subjective health of the respondent.
21
madrasa enrolment from a within household perspective. However, these issues are left
for future research. Nonetheless in the empirical analysis we control for child age and
gender at the household level in order to acknowledge the relevance of demographic
factors.
Results
Table 5 presents the regression results following two different regression models, namely,
OLS and fractional logit. Two specifications are presented for each model. The first is a
parsimonious specification whilst the second additionally controls for village
characteristics. The main results are as follows.
Income effect: Madrasahs charge lower fees and hence may attract children from poorer
households. Households may be unable to afford admission fees, textbooks, transport and
private tuition required to attend government recognised schools and hence may often
find that their only option is a registered madrasah. Indeed in our theoretical model in
section 3, the effect of parental income (y) on the fraction of children sent to madrasahs is
negative.
To test for this possibility, we included 2 dummies which capture the economic status of
the household. For each household, we first compute the total food and non-food
expenditure. We then work out per capita monthly household expenditure, and apply adult
equivalence correction to account for scale economies arising due to differences in
household demographic structure. On the basis of this variable, we calculate expenditure
quartiles – the 2 dummies in our empirical specification correspond to the first 2 quartiles
leaving the top 2 quartiles as the base category. If household poverty is a valid
explanation for madrasah enrolment, we would expect positively signed coefficients on
dummies corresponding to lower quartiles. Indeed there is some evidence in favour of the
poverty hypothesis. The coefficient on the variable “Household expenditure: 2nd quartile”
is always significant in all regressions (see Table 5). This is also true for the variable
“Household expenditure: 1st quartile”. When other proxies of household poverty are
examined, the inverse relationship between household income and madrasah attendance is
also confirmed. Madrasah enrolment is also found to be negatively associated with
electricity connection to the household’s home.
22
Table 5: Household level estimation of the determinants of madrasa enrolment
(Dependent variable: Proportion of enrolled school children who attended
madrasahs)
OLS
Average child characteristics
Fraction of children being female
Average age of children in the household
Parental background
Mother education: primary completed
Mother education: secondary completed
Father education: primary completed
Father education: secondary completed
(1)
(2)
0.012
0.011
(0.49)
(0.44)
(0.52)
(0.51)
-0.001
-0.004
-0.001
-0.003
(0.19)
(0.56)
(0.15)
(0.53)
-0.007
0.004
-0.012
-0.006
(0.15)
(0.07)
(0.28)
(0.14)
-0.028
-0.004
-0.033
-0.005
(0.42)
(0.05)
(0.45)
(0.08)
0.019
0.055
(0.32)
(0.9)
Parental religiosity: Whether parents pray daily
Household expenditure: 2nd quartile
Household has electricity
Distance of the household from: Primary school
Distance of the household from: Secondary
madrasa
0.012
0.012
0.012
(0.24)
0.057
0.052
(0.95)
0.077
0.072
(1.31)
(1.17)
(1.48)
0
-0.001
-0.0003
-0.001
(0.21)
(0.59)
(0.22)
(0.81)
0.002
0.004
0.002
0.003
(0.75)
(1.27)
(0.65)
(1.29)
0.083
0.09
(3.31)**
(3.48)**
0.079
(3.23)**
0.084
(3.36)**
0
0.001
0.0009
(0.04)
(0.36)
(0.25)
(0.4)
0.057
(2.00)*
0.058
(1.97)*
0.061
(2.19)*
0.063
(2.21)*
0.055
(1.91)+
0.065
(2.33)*
-0.085
(3.52)**
0.057
(1.93)+
0.069
(2.41)*
-0.083
(3.22)**
0.001
0.0003
0.0001
-0.087
-0.087
(3.45)**
(3.28)**
0
0
(0.65)
(0.17)
(1.03)
(0.35)
0.004
0.004
(4.22)**
(3.80)**
-0.001
-0.001
0.0037
(4.02)**
-0.0014
0.0039
(4.06)**
-0.0016
(0.97)
(1.02)
(1)
(1.13)
-0.002
-0.002
-0.0015
-0.0022
School quality and availability
Distance of the household from: Sub-district
headquarters
Distance of the household from: Secondary
school
(2)
0.057
Parental religiosity: the frequency of reciting the
Quran
Household attributes
Household expenditure: 1st quartile
(1)
(1.06)
Father's age
Mother's age
Fractional Logit
23
(1.58)
(1.96)+
-1.44
(2.11)*
Distance of the household from: Primary
madrasa
-0.002
-0.001
-0.0018
-0.0015
(1.22)
-0.033
(1.56)
(1.27)
log of mean Maths score of local madrasas
(1.5)
0.002
0.0073
-0.025
(0.07)
-0.059
(0.93)
-0.051
(0.23)
(0.67)
log of mean Maths score of local schools
(0.86)
(0.69)
(1.12)
(0.74)
Number of madrasas as a fraction of all
secondary schools
Number of madrasas as a fraction of all primary
schools
-0.076
0.256
0.408
(1.85)+
(2.64)**
0.285
(2.04)*
0.438
(2.73)**
-0.102
-0.164
-0.228
-0.299
(0.34)
(0.5)
(0.92)
(1.07)
Community characteristics
Village has electricity
Village has pucca(paved surface) road
Village: number of mosques
Village has NGO educational activity
0.056
0.043
(1.03)
(0.78)
-0.031
-0.03
(1.07)
(1.11)
-0.009
-0.009
(0.78)
(0.85)
-0.076
-0.079
(2.18)*
-0.051
(3.40)**
(2.06)*
Village: mean education of the population
-0.053
(3.74)**
Village: proportion of the population being nonMuslim
Constant
-0.05
-0.093
-0.069
(0.66)
(0.5)
-0.001
0.149
-3.296
-2.336
0
(0.63)
(2.18)*
(1.39)
Observations
1212
1163
1212
1163
Adj. R-squared
0.08
0.09
--
--
Log likelihood
--
--
-549.15
-523.64
chi2
--
--
126.35
140.14
Note: (a) + significant at 10%; * significant at 5%; ** significant at 1%. (b) Results based on sample of all
children aged 11-18 years who attended secondary education. (c) Base-category for parental education is
“less than 5 years of schooling”. (d) Standard errors are robust to clustering-effect at village level. (e) The
variable “Parental religiosity: the frequency of reciting the Quran” ranges between 1 (recites everyday) and
5 (recites once-twice in a year).
Religious preference of the household: We include two proxies of parental religiosity 21 –
“whether prays daily” and “the frequency of reciting the Quran”. In all specifications,
daily prayer is positively correlated with the proportion of children enrolled in madrasah.
21
Unlike in the theoretical model where parental religiosity is assumed to be a continuous index, for
regression purposes, religiosity is a binary variable (1 if the household head prays daily; 0 otherwise). Given
our current dataset, it is not possible to distinguish between different degrees of parental religiosity.
24
This correlation could be dismissed on the ground that religiosity is simply a proxy for
household poverty. However, the regression model already includes detailed control for
household economic conditions. Neither can the influence of parental religiosity be
attributed to the influence of local poverty 22 since the correlation survives even in
specification where detailed control for village characteristics is present. Nonetheless, the
evidence presented here should be treated as descriptive instead of causal: whether or not
the parents recite the Quran could be also affected by the school their children attend.
Access to schools: Madrasah enrolment is positively correlated with distance of the
household from non-religious secondary schools. The opposite is true for distance from
secondary madrasahs. This finding is consistent with the positive and statistically
significant correlation obtained between madrasah enrolment and relative availability of
madrasahs (as a fraction of all the secondary schools in the sample cluster).
School quality: This is measured using data on average test scores of a sample of grade 8
students in all the secondary non-religious schools and madrasahs in sample cluster
(namely union) 23. Irrespective of the regression specification used, the correlation
between madrasah enrolment and average quality of madrasahs in the union is
insignificant. The same result is obtained for the average quality of schools in the union.
This finding remains unchanged irrespective of whether we use mathematics or English
test scores of students to measure quality.
Community (village-specific) influence: The second specification of each regression
model in Table 5included 6 correlates of village level development – village with
electricity, average educational attainment of the (adult) villagers, number of mosques in
the village, presence of pucca (paved surface) road in the village, proportion of nonMuslim population and village with NGO-run education programs. Of all the above
22
Much of the household preference effect can be conflated by supply-side variables such as availability of
formal schools and level of development of the village of development in the village of residence. It is
because children from religious households may also live in villages that are underprovided in terms of the
supply of various public goods.
23
We use the log of average test scores which was collected from an accompanying survey of all
educational institutions in the 48 sample clusters. Given that the quality variable is measured as an
aggregate average, this serves as an exogenous proxy for the quality of secondary schools/madrasahs in the
sample clusters.
25
variables, only the average education of the villagers enters the regression function
significantly and is positively associated with madrasah enrolment.
Robustness of the results: It should be noted that our conclusions obtained from
fractional logit regression estimates presented in Table 5 remain unchanged when we
pursue alternative estimation strategy, namely, OLS regressions. Lastly, a significant part
of the recent growth in madrasah enrolment reflects the general increase in secondary
school participation by girls across Bangladesh. Whilst registered madrasahs were
historically a preserve for boys, they started admitting girls following the introduction of
the “Female Secondary School Stipend Program” (FSSSP) (Asadullah and Chaudhury
2009b). To this end, we re-estimated the regression model restricting data to the sample of
male children (results not reported but available from the authors). However, our results
remained unchanged. This finding is interesting because it could be argued that old-age
transfer, if valid, is more relevant in predicting enrolment decisions for boys. Our analysis
therefore suggests that parents’ perception of the old-age support from educated daughters
in rural Bangladesh may be changing, an issue that needs to be examined in future
research.
6. Conclusion and policy implications
In this paper we have examined the choice of registered madrasah education in rural
Bangladesh in the context of a two-period school choice model. Our empirical analysis is
based on a unique set of data that combines survey information on parents' choices and
socio-economic characteristics with survey data on all schools and madrasahs that are de
facto available for each family in the study area. We then assessed the choices made in
terms of the characteristics of families, schools and communities.
Our findings indicate that economic considerations such as income and schooling cost
dominate in the decision to educate children in registered secondary madrasahs. Poorer
households, and those with worse access (and hence higher travel cost) to non-religious
schools tend to enrol a larger proportion of their children in madrasahs. Although
madrasah enrolment varies positively with household religiosity, we find little evidence to
suggest that the average rural Bangladeshi household possesses an innate preference for
madrasah education which is insensitive to economic factors.
26
Our findings have important implications for education policy design. Both the theoretical
model and the empirical results suggest that education policies such as increasing the
supply of non-religious mainstream schools in rural locations can help raise enrolment in
these schools. The theoretical model also supports policies that boost the quality of nonreligious schools. However, in contrast to the theory, we find little empirical evidence in
support of policies that improve relative quality of non-religious schools. We in fact find
that religious school demand does not respond to the average quality of non-religious
schools and madrasahs in the locality. This finding can be explained by the fact that
education quality remains low across all types of institutions in rural Bangladesh
(Asadullah, Chaudhury and Dar, 2007) 24.The learning premium associated with
mainstream non-religious schools may not be sufficiently large so as to make a significant
difference for household school choice at present. However, it is entirely possible, and
indeed likely, that further investment in non-religious school quality (for example through
curriculum changes or better teacher training) will favourably alter parents’ perception of
the premium associated with non-religious schools. Considering the fact that nonreligious schools already enjoy a resource advantage over their religious counterparts, this
may turn out to be a significant policy challenge.
Alongside improving the availability and quality of non-religious schools, it is also
important to address the quality issue in recognised madrasahs. Religious schooling
serves an important purpose in rural Bangladesh; it not only meets the needs of parents
who prefer a more religious curriculum, but in some communities, it is the only socially
acceptable form of schooling for adolescent girls. Thus simply increasing the supply of
non-religious schools would not in fact result in an increase in school enrolment for girls
in certain areas. Therefore, the Madrasah Education Board needs to focus on modernising
the madrasah curriculum as well as on teacher training and recruitment to make sure that
madrasah attendance does not unduly reduce the educational attainment of those for
whom madrasahs may be the only choice.
The cost of schooling has been shown to be an important factor in the household decision
making process. Thus financial transfers (such as stipends or school vouchers) may be
24
Other developing country studies that looked into relative quality of faith schools also obtained similar
results. For instance, see Backiny-Yetna and Wodon (2009) for the Democratic Republic of Congo.
27
used to make non-religious schools more affordable to certain households. However, this
increased enrolment in non-religious schools will have positive implications for human
capital accumulation only if there are concurrent policies to further improve the quality of
these schools. Finally, since household income is a major determinant of school choice, it
is likely that policies aimed at reducing poverty (some of which may already be in place)
will have the additional impact of increasing enrolment in non-religious, mainstream
schools.
In this study, we have abstracted away from the issue of demand for madrasahs which are
not recognised by the state. These madrasahs have a very small market share, and their
demand remains less understood considering the fact that they impart little marketable
skills. In this regard, non-market explanations can be important. First, in villages where
governments and economies function poorly, unrecognised madrasahs can become major
suppliers of social services, political action, and coercive force 25. Unrecognised
madrasahs are also a potential proxy for participation in religious organizations which can
help offset the negative effects of growing up in a disadvantaged environment 26.
Households secure their membership by sending at least one child in a madrasah.
Secondly, demand for madrasah education can be also motivated by marriage-related
considerations: Madrasah education can reinforce traditional values in girls which may
have substantial return in the marriage market.
Therefore, a natural direction for future research on religious school choice in rural
Bangladesh is to investigate the demand for these unrecognised madrasahs. From a
theoretical perspective, the model in this paper can be extended to incorporate not just
religious preferences of the household, but also the influence of social norms and other
non-market considerations. We can also further explore parents’ decisions regarding
exactly which children are picked to attend madrasahs, for example, based on their ability
or gender.
25
See Iannaccone and Berman (2006) for a similar argument.
Dehejia, DeLeiere, Luttmer and Mitchell (2007) report buffering effects of religious organizations on
youth outcomes in the US.
26
28
References
Abdalla, Raisuddin and Hussein (2004) Bangladesh Educational Assessment: Pre-primary
and Primary Madrasah Education Bangladesh. Dhaka: The United States Agency for
International Development (USAID).
Ahmed, A. M. (2009) “Are religious people more prosocial? A quasi-experimental study
with madrasah pupils in a rural community in India,” Journal for the Scientific
Study of Religion, 48, pp. 368-374.
Alderman, Harold, Peter Orazem and Elizabeth M. Paterno (2001) “School Quality,
School Cost, and the Public/Private School Choices of Low-Income Households in
Pakistan,” Journal of Human Resources, 36, (2), pp. 304-326.
Andrabi, Tahir, Jishnu Das, Asim Ijaz Khwaja and Tristan Zajonc (2006) “Religious
School Enrollment in Pakistan: A Look at the Data,” Comparative Education
Review, 50(3), pp. 446-477.
Asadullah, M. Niaz (2009) "Returns to Private and Public Education in Bangladesh and
Pakistan: A Comparative Analysis", Journal of Asian Economics, 20(1), pp. 77-86.
Asadullah, M. Niaz and Chaudhury, N. and Dar, A. (2007) "Student Achievement
Conditioned Upon School Selection: Religious and Secular Secondary School
Quality in Bangladesh", Economics of Education Review, 26(6), pp. 648-659.
Asadullah, M. Niaz and Chaudhury, N. (2009) "Holy Alliances: Islamic high schools,
public subsidies, and female schooling in Bangladesh," Education Economics,
17(3), pp. 377-394.
Asadullah, M. Niaz and Chaudhury, N. (2010) "Religious schools, social values and
economic attitudes: Evidence from Bangladesh", World Development, 38(2), pp.
205-217.
Backiny-Yetna, Prospere and Wodon, Quentin (2009a) "Comparing the Performance of
Faith-Based and Government Schools in the Democratic Republic of Congo,"
MPRA Paper 16463, University Library of Munich, Germany.
Backiny-Yetna, Prospere and Wodon, Quentin (2009b) "Comparing the Private Cost of
Education at Public, Private, and Faith-Based Schools in Cameroon," MPRA Paper
16464, University Library of Munich, Germany.
BANBEIS (2006) Secondary School Key Educational Indicators prepared by the
Bangladesh Bureau of Educational Information and Statistics. Accessed on
15/10/2006 from website
29
http://www.banbeis.gov.bd/db_bb/secondary_education_1.htm
CAMPE
(2005)
"The
State
of
Secondary
Education
Progress
and
Challenges".http://www.campebd.org/content/EW_2005.htm
Chang, Wen-Chun (2005) “Religious Giving, Non-religious Giving, and After-life
Consumption”, The B.E. Journal of Economic Analysis & Policy.
Cohen-Zada, D. (2006) “Preserving religious identity through education: Economic
analysis and evidence from the United States”, Journal of Urban Economics, 60,
pp. 372–398.
Cohen-Zada, D. and Justman, M. (2005) “The religious factor in private education”,
Journal of Urban Economics, 57, pp. 391–418.
Cohen-Zada, D. and Sander, W. (2008) “Religion, religiosity, and private school choice:
Implications for estimating the effectiveness of private schools”, Journal of Urban
Economics, 64(1), pp. 85-100.
Dehejia, Rajeev, Thomas DeLeiere, Erzo F.P. Luttmer and Joshua Mitchell (2007)“The
Role of Religious and Social Organizations in the Lives of Disadvantaged Youth”,
NBER Working Papers 13369, National Bureau of Economic Research, Inc.
Evans, A. (2008) "Madrasah Education: Necessity or Rational Choice?" Harvard
International Review, Vol. 30(3).
Garg, Ashish and Jonathan Morduch (1998) "Sibling rivalry and the gender gap: Evidence
from child health outcomes in Ghana," Journal of Population Economics, 11(4), pp.
471-493.
Iannaccone, Laurence and Berman, Eli (2006) “Religious extremism: The good, the bad,
and the deadly”, Public Choice, 128(1), pp. 109-129.
Long, J. and Toma, E. (1998) “The determinants of private school attendance”, Review of
Economics and Statistics, 70, pp. 458–479.
Neal, D (1997) “The Effects of Catholic Secondary Schooling on Educational
Achievement”, Journal of Labor Economics, 15(1), pp. 98-123.
Newhouse, D., and Beegle, K. (2006) “The effect of school type on academic
achievement: Evidence from Indonesia”Journal of Human Resources, 41 (3), pp.
529-557.
Papke and Wooldridge (1996) "Econometric methods for fractional response variables
with an application to 401(k) plan participation rates" Journal of Applied
Econometrics, 11(6), pp. 619-632.
30
Sander, William (2005) “Catholics and Catholic schooling”, Education Economics13,pp.
257–268.
Sander, William (2005) “Religion, Religiosity, and Private Schools,” Catholic Education:
A Journal of Inquiry and Practice,” 9(1), pp. 7-21.
Sattar, Abdus (2004) Bangladesh madrasa shikhkha o samajjibone tar provab (Madrasa
education in Bangladesh and indications in social life). Dhaka: Islamic Foundation.
Wagner, J (2001) “A note on the firm size-export relationship”, Small Business
Economics17, pp. 229–337.
Wodon, Quentin and Yvonne Ying (2009) “Literacy and Numeracy in Faith-Based and
Government Schools in Sierra Leone”.
Wooldridge, William (2002) Econometric Analysis of Cross Section and Panel Data,
Cambridge, MA: MIT Press.
World Bank (2010) “Secondary School Madrasas in Bangladesh: Incidence, Quality, and
Implications for Reform”, Bangladesh Development Studies (BDS) Series no 27.
31
Appendix 1
𝑎
The function 𝑣 �𝑍 � may be generically illustrated as follows:
𝑎
𝑃
𝑣 �𝑍 �
𝑃
1
(𝑎/𝑍𝑃 )
32
𝑎
�𝑍 �
𝑃
Appendix Table 1: Summary statistics
Sample of
households with at
least 1 secondary
school enrolled
children
Sample of children
enrolled in a
secondary school
Sample of children
enrolled in a
secondary madrasah
Mean
Std.
Dev.
Mean
Std.
Dev.
Mean
Madrasah attendance
0.21
0.39
0
--
1
--
Child being female (mean)
0.59
0.46
0.59
0.49
.59
0.49
Child's age (mean)
15.37
1.57
15.41
1.79
15.37
1.78
Mother’s education: primary completed
0.21
0.41
0.21
0.41
0.25
0.44
Mother’s education: secondary completed
0.03
0.18
0.03
0.18
0.03
0.16
Father’s education: primary completed
0.81
0.39
0.81
0.39
0.77
0.42
Father’s education: secondary completed
0.11
0.31
0.11
0.31
0.14
0.35
Father's age
42.29
7.07
42.41
7.04
41.82
7.43
Mother's age
40.11
5.29
40.03
5.52
40.50
4.87
Per capita monthly expenditure in Taka
2317.69
1066.17
2403.17
1143.08
2244.38
936.56
(in USD*)
(34.08)
(15.67)
(35.34)
(16.81)
(33.00)
(13.77)
Household expenditure: 1st quartile
0.31
0.46
0.28
0.45
0.32
0.47
Household expenditure: 2nd quartile
0.28
0.45
0.28
0.45
0.31
0.47
Parental religiosity: Whether parents pray daily
0.57
0.50
0.56
0.50
0.70
0.46
Parental religiosity: the frequency of reciting the Quran
Distance of the household from: Sub-district headq (in
minutes)
Distance of the household from: Secondary school (in
minutes)
Distance of the household from: Primary school (in
minutes)
Distance of the household from: Secondary madrasa (in
minutes)
Distance of the household from: Primary madrasa (in
minutes)
4.28
3.32
4.41
3.32
4.55
3.44
44.70
30.02
44.11
29.63
48.95
31.74
20.49
13.70
19.84
12.96
22.57
15.76
11.52
10.28
11.28
9.89
11.12
8.75
21.81
14.99
22.06
14.99
19.78
14.45
18.17
13.59
18.46
13.71
15.57
11.43
Household has electricity
0.59
0.49
0.63
0.48
0.53
0.50
Village has electricity
0.93
0.25
0.93
0.25
0.95
0.20
Village has pucca (paved surface) road
0.63
0.48
0.65
0.48
0.55
0.50
Village: # of mosques
2.37
1.52
2.30
1.50
2.53
1.52
Village has NGO educational activity
0.88
0.33
0.88
0.32
0.85
0.35
Village: mean education of the population
4.71
1.12
4.76
1.13
4.80
1.07
Village: fraction of non-Muslim population
0.06
0.10
0.06
0.11
0.06
0.08
Mean English score of local madrasas (in logs)
1.39
0.90
1.36
0.92
1.57
0.84
Mean English score of local schools (in logs)
4.86
1.41
4.93
1.44
4.66
1.25
Mean Maths score of local madrasas (in logs)
2.14
1.19
2.09
1.21
2.46
1.08
Mean Maths score of local schools (in logs)
5.88
1.62
5.92
1.64
5.88
1.51
N
1229
* 1 USD was equivalent to 68 taka at the time of the field work.
33
1264
343
Std. Dev.